Precision of dual-photon absorptiometry (DPA) measurements was determined in a lumbar spine phantom and in humans. Approximately half of the measurements were made before and half after a 153gadolinium source change. The phantom was measured with different amounts of acrylic, which simulates human soft tissue, in order to evaluate the influence of body thickness on bone mineral density (BMD). Results of scans analyzed with two software versions from Lunar Radiation Corp., the widely used 08B and a prototype 08C, are compared. DPA with a cold source significantly overestimated BMD in the phantom in the presence of large amounts (more than 25 cm) of soft tissue equivalent with version 08B but not with the newer version 08C. Similarly, in nine subjects, there was a significant decrease in spine BMD after a source change when scans were analyzed with version 08B (mean difference 0.026 g/cm2, P = 0.002) but not with 08C (0.01 g/cm2, P = 0.234). No systematic effect of source change on femoral BMD measurements was observed. The SD of the mean difference of two measurements of the nine subjects was 0.019 g/cm2 (1.6% of the mean value) for the spine with software version 08B and 0.024 g/cm2 (2.0%) with version 08C, 0.03 g/cm2 (3.3%) for the femur neck, 0.03 g/cm2 (4.0%) for the greater trochanter, and 0.04 g/cm2 (4.9%) for Ward's triangle region of the proximal femur. The spine phantom was scanned on two other commercial bone densitometers in order to assess inter-instrument variation.(ABSTRACT TRUNCATED AT 250 WORDS)